Workslop 📎

The office promises of AI are dazzling. Less admin, fewer meetings, cleaner drafts, faster summaries. The end of the blank page.

Knowledge work has always contained a degree of ritualised nonsense. AI promises to do away with much of it. That’s great. But we also face a tsunami of workslop: AI-generated work that appears useful but lacks substance, contains inaccurecies, or needs someone else to correct or redo it entirely. One person saves ten minutes producing a first pass, then two more spend half an hour working out what the hell it means.

Clock up

It’s not the first time work has confused activity with productivity. Engineer Frederick Winslow Taylor was, more than a century ago, timing workers with a stopwatch, breaking their toil into measurable units. Taylorism made many factories more efficient, sure. But it helped give rise to the managerial fantasy that work becomes more real the more it’s measured.

In the modern era, the photocopier made duplication cheap, so documents multiplied. Email made distribution free, so everyone got copied in. AI makes composition cheap, so the risk is that work-adjacent content multiplies too. This is Goodhart’s Law on steroids. When a measure becomes a target, it ceases to be a good measure.

External affairs

In one survey, 40% of 1,150 US employees said they had received workslop in the past month, and it’s estimated to make up about 15% of workplace content.

Offices have always had some bad output. The difference is, it can now be produced faster and distributed with more confidence than ever before. Having said that, AI gains are real, to some extent. 

In one survey, ChatGPT Enterprise users attributed 40-60 minutes saved per day to AI use, while Microsoft’s (MSFT) analysis of 50,000 Copilot-enabled Word users found an average difference of seven minutes per output. Across a large company, that is not trivial. But saving time and creating value aren’t always the same thing. Worth remembering, of course, that Microsoft is not a neutral observer of AI productivity gains.

There’s a word for costs that get pushed elsewhere: externalities. A factory producing clothing pollutes a river. A driver on her way to work adds to congestion. Workslop is an office externality. The person producing it gets the benefit. The people downstream absorb the cost.

Calendar invite

Organisations are already very good at turning time into work-shaped matter. C. Northcote Parkinson argued in the 1950s that “work expands so as to fill the time available for its completion”. Parkinson’s famous example was the British Admiralty. 

In 1914, the Royal Navy had 62 capital ships and around 146,000 officers and men. By 1928, after the First World War and naval retrenchment, the number of capital ships had fallen to 20, and the number of officers and men to about 100,000. 

But, over the same period, Admiralty officials rose from roughly 2,000 to 3,569. Dockyard officials and clerks rose too, from 3,249 to 4,558. Parkinson dryly referred to this as “a magnificent navy on land”. The fleet had shrunk while the paperwork had ballooned.

Power to the point

As well as making work for work’s sake, technology can give unfinished thinking a sheen of completeness and competence. Edward Tufte’s critique of PowerPoint argued the presentation software, which, by the late 90s, was dominant across offices, encourages hierarchy, thin evidence, and bullet-point compression.

His most alarming example was NASA’s handling of information before the Columbia shuttle disaster, claiming crucial engineering concerns were buried inside dense, knotty PowerPoint slides.

In one slide, Tufte argued, the most important caveats were buried in small text and lower-level bullets, while the headline gave a more reassuring impression than the evidence justified. The Columbia Accident Investigation Board later included Tufte’s PowerPoint analysis in its report, as part of a wider critique of NASA’s communication culture. PowerPoint had not caused the disaster. But the format, built to make information presentable, had instead obscured the salient evidence.

The medium is the memo

Media theorist Marshall McLuhan, prescient fellow that he was, had something to say on the effect technology can have on our way of thinking and creating. His mantra, the medium is the message, suggests the way we transmit a message changes the messaging. 

A slide deck and a chatbot do not just package the same thought differently. They push us to interpret and conceive in different ways. The 7-inch record helped standardise the modern pop single, for example. The memo encouraged brevity and clarity. PowerPoint turned ideas into bullets. AI fashions them into fluent prose. 

A recent essay makes the case AI is not just a productivity tool or labour market disruptor but a technology that may change how people think, much as literacy changed the brain over centuries. We may be trading deep thinking for quick answers, with costs to memory, judgement and confidence, and it suggests the winners may be those who can still think… er, without the machine. Where was I? Oh, yes.

Trust issues

Research shows people who use AI can be judged as lazier, less competent, less diligent, less trustworthy. And less moral! In some cases, saying you used AI can erode trust. So the modern wagie faces a neat little trap: do not use AI and you look slow; use it and you may look lazy; disclose it and you look suspect.

Generative AI shifts effort from ‘thinking by doing’ to ‘choosing from outputs’. That sounds efficient, because it is. But it also creates a skills problem. Doing the work teaches you what good looks like. 

Choosing between three AI-generated versions teaches something else: taste, editing, delegation, prompt management, perhaps. Those are useful skills, but they depend to some extent on prior mastery. If you have not struggled through the thing yourself, it is harder to know whether the machine has done it well.

Interns in the coal mine

And it does look like AI may be replacing the work that teaches the beginner how to become an expert. Juniors have always been responsible for a certain amount of drudgery, and not just making the coffees. Much of it is boring. But boring is often where judgement is cultivated and nurtured. Take away the dull bits, as AI so masterfully does, and you take away the repetition that turns rules into instinct and instinct into wisdom.

The gloomy outlook is some white-collar careers increasingly find the bottom rungs missing. Fine for those already halfway up. Less fine for the wave of graduates entering the workforce.

Although large-scale studies in Denmark and the US have not found significant aggregate effects of AI on unemployment or job openings, payroll data for workers aged 22-25 in highly AI-exposed jobs showed a fall of about 13% compared with less-exposed roles. UK evidence points in the same direction: McKinsey has warned that if entry-level hiring continues to slow in AI-exposed professional roles, companies risk damaging the future talent pipeline.

Learn-to mode

The labour market impact of AI may depend on how occupations are restructured and whether demand rises for the remaining work. ATMs and bank tellers have worked peacefully side by side for decades now. The machines handle the cash, while we manage the soft, squishy skills that humans are still bestest at. 

Spreadsheets automated calculation but increased demand for modelling and scenario planning. AI could do something similar if lawyers spend less time on document review and more on judgement and negotiation, or developers spend less time writing boilerplate and more on testing and product decisions.

The future may be AI-shaped, but it will also need a healthy dose of human nous, of frank and honest judgement. That slow, slightly annoying habit of asking questions, determining whether the work is legit useful, accurate, and complete.

Busy work

The office itself was one of the original productivity technologies. As it grew alongside telephones and typewriters, filing cabinets and carbon paper, it became a machine for coordinating knowledge work. It made work visible. The memo, the meeting, were evidence that something productive was happening. 

Maybe the problem is knowledge work is often managed as if it were factory work. Office workers are not screwing caps onto toothpaste tubes. They’re interpreting messy information, spotting patterns, writing, deciding, persuading, prioritising, and occasionally pretending to listen. Workslop is becoming an increasing share of that material. The need for evidence can make genuine productivity hard to measure. Work expands to fill the calendar.

Plenty of office faff deserves to die. But some difficulty is how judgement forms. The trick, as AI settles into office work, will be teasing apart useless friction from useful struggle.

‍

**Would you like me to do another pass?**